For drones that are being introduced to various areas, one of the expected uses in the disaster prevention field is in the disaster prevention field.By gathering information from the sky from the sky to collect information, the overall image of the damage is grasped, the victims squirted in the incident / accident are confirmed as soon as possible, or the communication infrastructure is revived to the affected areas as a "flying base station".It is assumed to be used such as making it, and has been put into practical use.
For example, in Tokyo alone, it is expected that the market for disaster prevention drone services will reach 300 million yen by 2030.These disaster prevention drones continue to evolve to help the victims faster and more reliably.One of these new research drones is from the German Frown Hoofer Research Organization.
The Frown Hoofer Research Organization is a research institute specializing in applying various science and technology, and has 75 research institutes and facilities throughout Germany.Each laboratory specializes in individual technologies such as chemistry, electronic engineering, and computer science, and the entire Frown Hoofer Research Organization costs about 2.8 billion euros (about 370 billion yen) per year.There is.
In June this year, one of the facilities that make up the research organization, the Institute of Communications, Information Processing, and Human Engineering (FRUNHOFER FKIE), has published research on drones at Acoustic Society of America, an international academic society of acoustic studies.It is no wonder that drone is now a machine controlled by advanced information processing, so that drones are now called "flying robots", so it is no wonder that a laboratory specializing in communication and information processing handles drones.But why was the research shown at acoustic conferences?
The answer is why this research is attracting attention.Developed drone demonstration videos are published on the site of the Society, so please check them out.
It's a short video, but if a woman in the square shouts, "Help me!", You can see that a dorone flying over the sky approaches the woman.This drone is not that anyone nearby is piloting.It is automatically approaching in response to women's screams.In other words, it is a drone that flew in the stricken area to collect and analyze audio, understand the existence of the victims as soon as possible.
As you can see from the demonstration, the concept of this search drone is very simple.However, if you respond to any voice or loud sound, you cannot judge the victim.It is not enough to simply judge that someone is looking for help, and you have to be able to accurately determine where you are.Therefore, drones also need "training", just as training rescue members and disaster rescue dogs.
The raw audio data collected by the drone contains unnecessary environmental sounds and the sounds created by the drone itself, so first remove those noise by filtering.From the data obtained, it is determined that the victim's voice and the presence or absence of sound are determined, but the familiar AI (artificial intelligence) is used there.
The affected person asks for help by various means to get the searcher to notice.However, it is not always the case that the victim uses.In some cases, you may be injured when you get caught in a disaster and lose your voice, and you may try to draw attention by hitting your hand or kicking something.In order to let AI learn the "pattern when humans call for help" in such voices and sounds, researchers record many types of sounds that are considered to be produced when humans fall into difficulties, and train them.Used as data.
Although it is not a drone training, research is already underway to extract specific types (patterns) from audio data to AI in various forms.For example, in a paper published by French researchers in 2016, one of the techniques to realize AI and a screaming and shouting of the data collected in the subway car.We have succeeded in extracting and classifying other audio.If data is accumulated in the future, it will be possible to further increase accuracy.
If the victim's voice or sound is detected, the next position will be determined.Here, a microphone array called a can (Crow's Nest Array) is used (a device that is composed of multiple microphones, used for applications such as identifying sound sources and extracting specific audio), and accurate the source of the occurrence.You can grasp.However, it is meaningless if it can not be used on a drone, no matter how high -performance it is, so this study is also undergoing a series of minority and lighter weight.
According to Fraunhofer Fkie's announcement, many experiments and verifications, including field tests, are being promoted, and good results have been obtained.It has already been detected since the audio has already reached a level that can identify the victim's position within a few seconds, and a prototype as a product has been created.In the near future, it may be useful in the field of natural disasters, or in urban areas, as a system that detects suspicious sounds such as screams and gunshots and expresses them on the scene.